1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2013 Gary Mills
  23  *
  24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  25  */
  26 
  27 /*
  28  * zlogin provides three types of login which allow users in the global
  29  * zone to access non-global zones.
  30  *
  31  * - "interactive login" is similar to rlogin(1); for example, the user could
  32  *   issue 'zlogin my-zone' or 'zlogin -e ^ -l me my-zone'.   The user is
  33  *   granted a new pty (which is then shoved into the zone), and an I/O
  34  *   loop between parent and child processes takes care of the interactive
  35  *   session.  In this mode, login(1) (and its -c option, which means
  36  *   "already authenticated") is employed to take care of the initialization
  37  *   of the user's session.
  38  *
  39  * - "non-interactive login" is similar to su(1M); the user could issue
  40  *   'zlogin my-zone ls -l' and the command would be run as specified.
  41  *   In this mode, zlogin sets up pipes as the communication channel, and
  42  *   'su' is used to do the login setup work.
  43  *
  44  * - "console login" is the equivalent to accessing the tip line for a
  45  *   zone.  For example, the user can issue 'zlogin -C my-zone'.
  46  *   In this mode, zlogin contacts the zoneadmd process via unix domain
  47  *   socket.  If zoneadmd is not running, it starts it.  This allows the
  48  *   console to be available anytime the zone is installed, regardless of
  49  *   whether it is running.
  50  */
  51 
  52 #include <sys/socket.h>
  53 #include <sys/termios.h>
  54 #include <sys/utsname.h>
  55 #include <sys/stat.h>
  56 #include <sys/types.h>
  57 #include <sys/contract/process.h>
  58 #include <sys/ctfs.h>
  59 #include <sys/brand.h>
  60 #include <sys/wait.h>
  61 #include <alloca.h>
  62 #include <assert.h>
  63 #include <ctype.h>
  64 #include <door.h>
  65 #include <errno.h>
  66 #include <nss_dbdefs.h>
  67 #include <poll.h>
  68 #include <priv.h>
  69 #include <pwd.h>
  70 #include <unistd.h>
  71 #include <utmpx.h>
  72 #include <sac.h>
  73 #include <signal.h>
  74 #include <stdarg.h>
  75 #include <stdio.h>
  76 #include <stdlib.h>
  77 #include <string.h>
  78 #include <strings.h>
  79 #include <stropts.h>
  80 #include <wait.h>
  81 #include <zone.h>
  82 #include <fcntl.h>
  83 #include <libdevinfo.h>
  84 #include <libintl.h>
  85 #include <locale.h>
  86 #include <libzonecfg.h>
  87 #include <libcontract.h>
  88 #include <libbrand.h>
  89 #include <auth_list.h>
  90 #include <auth_attr.h>
  91 #include <secdb.h>
  92 
  93 #ifdef  LOGNAME_MAX_ILLUMOS
  94 #define _LOGNAME_MAX    LOGNAME_MAX_ILLUMOS
  95 #else /* LOGNAME_MAX_ILLUMOS */
  96 #define _LOGNAME_MAX    LOGNAME_MAX
  97 #endif /* LOGNAME_MAX_ILLUMOS */
  98 
  99 static int masterfd;
 100 static struct termios save_termios;
 101 static struct termios effective_termios;
 102 static int save_fd;
 103 static struct winsize winsize;
 104 static volatile int dead;
 105 static volatile pid_t child_pid = -1;
 106 static int interactive = 0;
 107 static priv_set_t *dropprivs;
 108 
 109 static int nocmdchar = 0;
 110 static int failsafe = 0;
 111 static char cmdchar = '~';
 112 
 113 static int pollerr = 0;
 114 
 115 static const char *pname;
 116 static char *username;
 117 
 118 /*
 119  * When forced_login is true, the user is not prompted
 120  * for an authentication password in the target zone.
 121  */
 122 static boolean_t forced_login = B_FALSE;
 123 
 124 #if !defined(TEXT_DOMAIN)               /* should be defined by cc -D */
 125 #define TEXT_DOMAIN     "SYS_TEST"      /* Use this only if it wasn't */
 126 #endif
 127 
 128 #define SUPATH  "/usr/bin/su"
 129 #define FAILSAFESHELL   "/sbin/sh"
 130 #define DEFAULTSHELL    "/sbin/sh"
 131 #define DEF_PATH        "/usr/sbin:/usr/bin"
 132 
 133 #define CLUSTER_BRAND_NAME      "cluster"
 134 
 135 /*
 136  * The ZLOGIN_BUFSIZ is larger than PIPE_BUF so we can be sure we're clearing
 137  * out the pipe when the child is exiting.  The ZLOGIN_RDBUFSIZ must be less
 138  * than ZLOGIN_BUFSIZ (because we share the buffer in doio).  This value is
 139  * also chosen in conjunction with the HI_WATER setting to make sure we
 140  * don't fill up the pipe.  We can write FIFOHIWAT (16k) into the pipe before
 141  * blocking.  By having ZLOGIN_RDBUFSIZ set to 1k and HI_WATER set to 8k, we
 142  * know we can always write a ZLOGIN_RDBUFSIZ chunk into the pipe when there
 143  * is less than HI_WATER data already in the pipe.
 144  */
 145 #define ZLOGIN_BUFSIZ   8192
 146 #define ZLOGIN_RDBUFSIZ 1024
 147 #define HI_WATER        8192
 148 
 149 /*
 150  * See canonify() below.  CANONIFY_LEN is the maximum length that a
 151  * "canonical" sequence will expand to (backslash, three octal digits, NUL).
 152  */
 153 #define CANONIFY_LEN 5
 154 
 155 static void
 156 usage(void)
 157 {
 158         (void) fprintf(stderr, gettext("usage: %s [ -CES ] [ -e cmdchar ] "
 159             "[-l user] zonename [command [args ...] ]\n"), pname);
 160         exit(2);
 161 }
 162 
 163 static const char *
 164 getpname(const char *arg0)
 165 {
 166         const char *p = strrchr(arg0, '/');
 167 
 168         if (p == NULL)
 169                 p = arg0;
 170         else
 171                 p++;
 172 
 173         pname = p;
 174         return (p);
 175 }
 176 
 177 static void
 178 zerror(const char *fmt, ...)
 179 {
 180         va_list alist;
 181 
 182         (void) fprintf(stderr, "%s: ", pname);
 183         va_start(alist, fmt);
 184         (void) vfprintf(stderr, fmt, alist);
 185         va_end(alist);
 186         (void) fprintf(stderr, "\n");
 187 }
 188 
 189 static void
 190 zperror(const char *str)
 191 {
 192         const char *estr;
 193 
 194         if ((estr = strerror(errno)) != NULL)
 195                 (void) fprintf(stderr, "%s: %s: %s\n", pname, str, estr);
 196         else
 197                 (void) fprintf(stderr, "%s: %s: errno %d\n", pname, str, errno);
 198 }
 199 
 200 /*
 201  * The first part of our privilege dropping scheme needs to be called before
 202  * fork(), since we must have it for security; we don't want to be surprised
 203  * later that we couldn't allocate the privset.
 204  */
 205 static int
 206 prefork_dropprivs()
 207 {
 208         if ((dropprivs = priv_allocset()) == NULL)
 209                 return (1);
 210 
 211         priv_basicset(dropprivs);
 212         (void) priv_delset(dropprivs, PRIV_PROC_INFO);
 213         (void) priv_delset(dropprivs, PRIV_PROC_FORK);
 214         (void) priv_delset(dropprivs, PRIV_PROC_EXEC);
 215         (void) priv_delset(dropprivs, PRIV_FILE_LINK_ANY);
 216 
 217         /*
 218          * We need to keep the basic privilege PROC_SESSION and all unknown
 219          * basic privileges as well as the privileges PROC_ZONE and
 220          * PROC_OWNER in order to query session information and
 221          * send signals.
 222          */
 223         if (interactive == 0) {
 224                 (void) priv_addset(dropprivs, PRIV_PROC_ZONE);
 225                 (void) priv_addset(dropprivs, PRIV_PROC_OWNER);
 226         } else {
 227                 (void) priv_delset(dropprivs, PRIV_PROC_SESSION);
 228         }
 229 
 230         return (0);
 231 }
 232 
 233 /*
 234  * The second part of the privilege drop.  We are paranoid about being attacked
 235  * by the zone, so we drop all privileges.  This should prevent a compromise
 236  * which gets us to fork(), exec(), symlink(), etc.
 237  */
 238 static void
 239 postfork_dropprivs()
 240 {
 241         if ((setppriv(PRIV_SET, PRIV_PERMITTED, dropprivs)) == -1) {
 242                 zperror(gettext("Warning: could not set permitted privileges"));
 243         }
 244         if ((setppriv(PRIV_SET, PRIV_LIMIT, dropprivs)) == -1) {
 245                 zperror(gettext("Warning: could not set limit privileges"));
 246         }
 247         if ((setppriv(PRIV_SET, PRIV_INHERITABLE, dropprivs)) == -1) {
 248                 zperror(gettext("Warning: could not set inheritable "
 249                     "privileges"));
 250         }
 251 }
 252 
 253 /*
 254  * Create the unix domain socket and call the zoneadmd server; handshake
 255  * with it to determine whether it will allow us to connect.
 256  */
 257 static int
 258 get_console_master(const char *zname)
 259 {
 260         int sockfd = -1;
 261         struct sockaddr_un servaddr;
 262         char clientid[MAXPATHLEN];
 263         char handshake[MAXPATHLEN], c;
 264         int msglen;
 265         int i = 0, err = 0;
 266 
 267         if ((sockfd = socket(AF_UNIX, SOCK_STREAM, 0)) == -1) {
 268                 zperror(gettext("could not create socket"));
 269                 return (-1);
 270         }
 271 
 272         bzero(&servaddr, sizeof (servaddr));
 273         servaddr.sun_family = AF_UNIX;
 274         (void) snprintf(servaddr.sun_path, sizeof (servaddr.sun_path),
 275             "%s/%s.console_sock", ZONES_TMPDIR, zname);
 276 
 277         if (connect(sockfd, (struct sockaddr *)&servaddr,
 278             sizeof (servaddr)) == -1) {
 279                 zperror(gettext("Could not connect to zone console"));
 280                 goto bad;
 281         }
 282         masterfd = sockfd;
 283 
 284         msglen = snprintf(clientid, sizeof (clientid), "IDENT %lu %s\n",
 285             getpid(), setlocale(LC_MESSAGES, NULL));
 286 
 287         if (msglen >= sizeof (clientid) || msglen < 0) {
 288                 zerror("protocol error");
 289                 goto bad;
 290         }
 291 
 292         if (write(masterfd, clientid, msglen) != msglen) {
 293                 zerror("protocol error");
 294                 goto bad;
 295         }
 296 
 297         bzero(handshake, sizeof (handshake));
 298 
 299         /*
 300          * Take care not to accumulate more than our fill, and leave room for
 301          * the NUL at the end.
 302          */
 303         while ((err = read(masterfd, &c, 1)) == 1) {
 304                 if (i >= (sizeof (handshake) - 1))
 305                         break;
 306                 if (c == '\n')
 307                         break;
 308                 handshake[i] = c;
 309                 i++;
 310         }
 311 
 312         /*
 313          * If something went wrong during the handshake we bail; perhaps
 314          * the server died off.
 315          */
 316         if (err == -1) {
 317                 zperror(gettext("Could not connect to zone console"));
 318                 goto bad;
 319         }
 320 
 321         if (strncmp(handshake, "OK", sizeof (handshake)) == 0)
 322                 return (0);
 323 
 324         zerror(gettext("Console is already in use by process ID %s."),
 325             handshake);
 326 bad:
 327         (void) close(sockfd);
 328         masterfd = -1;
 329         return (-1);
 330 }
 331 
 332 
 333 /*
 334  * Routines to handle pty creation upon zone entry and to shuttle I/O back
 335  * and forth between the two terminals.  We also compute and store the
 336  * name of the slave terminal associated with the master side.
 337  */
 338 static int
 339 get_master_pty()
 340 {
 341         if ((masterfd = open("/dev/ptmx", O_RDWR|O_NONBLOCK)) < 0) {
 342                 zperror(gettext("failed to obtain a pseudo-tty"));
 343                 return (-1);
 344         }
 345         if (tcgetattr(STDIN_FILENO, &save_termios) == -1) {
 346                 zperror(gettext("failed to get terminal settings from stdin"));
 347                 return (-1);
 348         }
 349         (void) ioctl(STDIN_FILENO, TIOCGWINSZ, (char *)&winsize);
 350 
 351         return (0);
 352 }
 353 
 354 /*
 355  * This is a bit tricky; normally a pts device will belong to the zone it
 356  * is granted to.  But in the case of "entering" a zone, we need to establish
 357  * the pty before entering the zone so that we can vector I/O to and from it
 358  * from the global zone.
 359  *
 360  * We use the zonept() call to let the ptm driver know what we are up to;
 361  * the only other hairy bit is the setting of zoneslavename (which happens
 362  * above, in get_master_pty()).
 363  */
 364 static int
 365 init_slave_pty(zoneid_t zoneid, char *devroot)
 366 {
 367         int slavefd = -1;
 368         char *slavename, zoneslavename[MAXPATHLEN];
 369 
 370         /*
 371          * Set slave permissions, zone the pts, then unlock it.
 372          */
 373         if (grantpt(masterfd) != 0) {
 374                 zperror(gettext("grantpt failed"));
 375                 return (-1);
 376         }
 377 
 378         if (unlockpt(masterfd) != 0) {
 379                 zperror(gettext("unlockpt failed"));
 380                 return (-1);
 381         }
 382 
 383         /*
 384          * We must open the slave side before zoning this pty; otherwise
 385          * the kernel would refuse us the open-- zoning a pty makes it
 386          * inaccessible to the global zone.  Note we are trying to open
 387          * the device node via the $ZONEROOT/dev path for this pty.
 388          *
 389          * Later we'll close the slave out when once we've opened it again
 390          * from within the target zone.  Blarg.
 391          */
 392         if ((slavename = ptsname(masterfd)) == NULL) {
 393                 zperror(gettext("failed to get name for pseudo-tty"));
 394                 return (-1);
 395         }
 396 
 397         (void) snprintf(zoneslavename, sizeof (zoneslavename), "%s%s",
 398             devroot, slavename);
 399 
 400         if ((slavefd = open(zoneslavename, O_RDWR)) < 0) {
 401                 zerror(gettext("failed to open %s: %s"), zoneslavename,
 402                     strerror(errno));
 403                 return (-1);
 404         }
 405 
 406         /*
 407          * Push hardware emulation (ptem), line discipline (ldterm),
 408          * and V7/4BSD/Xenix compatibility (ttcompat) modules.
 409          */
 410         if (ioctl(slavefd, I_PUSH, "ptem") == -1) {
 411                 zperror(gettext("failed to push ptem module"));
 412                 if (!failsafe)
 413                         goto bad;
 414         }
 415 
 416         /*
 417          * Anchor the stream to prevent malicious I_POPs; we prefer to do
 418          * this prior to entering the zone so that we can detect any errors
 419          * early, and so that we can set the anchor from the global zone.
 420          */
 421         if (ioctl(slavefd, I_ANCHOR) == -1) {
 422                 zperror(gettext("failed to set stream anchor"));
 423                 if (!failsafe)
 424                         goto bad;
 425         }
 426 
 427         if (ioctl(slavefd, I_PUSH, "ldterm") == -1) {
 428                 zperror(gettext("failed to push ldterm module"));
 429                 if (!failsafe)
 430                         goto bad;
 431         }
 432         if (ioctl(slavefd, I_PUSH, "ttcompat") == -1) {
 433                 zperror(gettext("failed to push ttcompat module"));
 434                 if (!failsafe)
 435                         goto bad;
 436         }
 437 
 438         /*
 439          * Propagate terminal settings from the external term to the new one.
 440          */
 441         if (tcsetattr(slavefd, TCSAFLUSH, &save_termios) == -1) {
 442                 zperror(gettext("failed to set terminal settings"));
 443                 if (!failsafe)
 444                         goto bad;
 445         }
 446         (void) ioctl(slavefd, TIOCSWINSZ, (char *)&winsize);
 447 
 448         if (zonept(masterfd, zoneid) != 0) {
 449                 zperror(gettext("could not set zoneid of pty"));
 450                 goto bad;
 451         }
 452 
 453         return (slavefd);
 454 
 455 bad:
 456         (void) close(slavefd);
 457         return (-1);
 458 }
 459 
 460 /*
 461  * Place terminal into raw mode.
 462  */
 463 static int
 464 set_tty_rawmode(int fd)
 465 {
 466         struct termios term;
 467         if (tcgetattr(fd, &term) < 0) {
 468                 zperror(gettext("failed to get user terminal settings"));
 469                 return (-1);
 470         }
 471 
 472         /* Stash for later, so we can revert back to previous mode */
 473         save_termios = term;
 474         save_fd = fd;
 475 
 476         /* disable 8->7 bit strip, start/stop, enable any char to restart */
 477         term.c_iflag &= ~(ISTRIP|IXON|IXANY);
 478         /* disable NL->CR, CR->NL, ignore CR, UPPER->lower */
 479         term.c_iflag &= ~(INLCR|ICRNL|IGNCR|IUCLC);
 480         /* disable output post-processing */
 481         term.c_oflag &= ~OPOST;
 482         /* disable canonical mode, signal chars, echo & extended functions */
 483         term.c_lflag &= ~(ICANON|ISIG|ECHO|IEXTEN);
 484 
 485         term.c_cc[VMIN] = 1;    /* byte-at-a-time */
 486         term.c_cc[VTIME] = 0;
 487 
 488         if (tcsetattr(STDIN_FILENO, TCSAFLUSH, &term)) {
 489                 zperror(gettext("failed to set user terminal to raw mode"));
 490                 return (-1);
 491         }
 492 
 493         /*
 494          * We need to know the value of VEOF so that we can properly process for
 495          * client-side ~<EOF>.  But we have obliterated VEOF in term,
 496          * because VMIN overloads the same array slot in non-canonical mode.
 497          * Stupid @&^%!
 498          *
 499          * So here we construct the "effective" termios from the current
 500          * terminal settings, and the corrected VEOF and VEOL settings.
 501          */
 502         if (tcgetattr(STDIN_FILENO, &effective_termios) < 0) {
 503                 zperror(gettext("failed to get user terminal settings"));
 504                 return (-1);
 505         }
 506         effective_termios.c_cc[VEOF] = save_termios.c_cc[VEOF];
 507         effective_termios.c_cc[VEOL] = save_termios.c_cc[VEOL];
 508 
 509         return (0);
 510 }
 511 
 512 /*
 513  * Copy terminal window size from our terminal to the pts.
 514  */
 515 /*ARGSUSED*/
 516 static void
 517 sigwinch(int s)
 518 {
 519         struct winsize ws;
 520 
 521         if (ioctl(0, TIOCGWINSZ, &ws) == 0)
 522                 (void) ioctl(masterfd, TIOCSWINSZ, &ws);
 523 }
 524 
 525 static volatile int close_on_sig = -1;
 526 
 527 static void
 528 /*ARGSUSED*/
 529 sigcld(int s)
 530 {
 531         int status;
 532         pid_t pid;
 533 
 534         /*
 535          * Peek at the exit status.  If this isn't the process we cared
 536          * about, then just reap it.
 537          */
 538         if ((pid = waitpid(child_pid, &status, WNOHANG|WNOWAIT)) != -1) {
 539                 if (pid == child_pid &&
 540                     (WIFEXITED(status) || WIFSIGNALED(status))) {
 541                         dead = 1;
 542                         if (close_on_sig != -1) {
 543                                 (void) write(close_on_sig, "a", 1);
 544                                 (void) close(close_on_sig);
 545                                 close_on_sig = -1;
 546                         }
 547                 } else {
 548                         (void) waitpid(pid, &status, WNOHANG);
 549                 }
 550         }
 551 }
 552 
 553 /*
 554  * Some signals (currently, SIGINT) must be forwarded on to the process
 555  * group of the child process.
 556  */
 557 static void
 558 sig_forward(int s)
 559 {
 560         if (child_pid != -1) {
 561                 pid_t pgid = getpgid(child_pid);
 562                 if (pgid != -1)
 563                         (void) sigsend(P_PGID, pgid, s);
 564         }
 565 }
 566 
 567 /*
 568  * reset terminal settings for global environment
 569  */
 570 static void
 571 reset_tty()
 572 {
 573         (void) tcsetattr(save_fd, TCSADRAIN, &save_termios);
 574 }
 575 
 576 /*
 577  * Convert character to printable representation, for display with locally
 578  * echoed command characters (like when we need to display ~^D)
 579  */
 580 static void
 581 canonify(char c, char *cc)
 582 {
 583         if (isprint(c)) {
 584                 cc[0] = c;
 585                 cc[1] = '\0';
 586         } else if (c >= 0 && c <= 31) {   /* ^@ through ^_ */
 587                 cc[0] = '^';
 588                 cc[1] = c + '@';
 589                 cc[2] = '\0';
 590         } else {
 591                 cc[0] = '\\';
 592                 cc[1] = ((c >> 6) & 7) + '0';
 593                 cc[2] = ((c >> 3) & 7) + '0';
 594                 cc[3] = (c & 7) + '0';
 595                 cc[4] = '\0';
 596         }
 597 }
 598 
 599 /*
 600  * process_user_input watches the input stream for the escape sequence for
 601  * 'quit' (by default, tilde-period).  Because we might be fed just one
 602  * keystroke at a time, state associated with the user input (are we at the
 603  * beginning of the line?  are we locally echoing the next character?) is
 604  * maintained by beginning_of_line and local_echo across calls to the routine.
 605  * If the write to outfd fails, we'll try to read from infd in an attempt
 606  * to prevent deadlock between the two processes.
 607  *
 608  * This routine returns -1 when the 'quit' escape sequence has been issued,
 609  * or an error is encountered, 1 if stdin is EOF, and 0 otherwise.
 610  */
 611 static int
 612 process_user_input(int outfd, int infd)
 613 {
 614         static boolean_t beginning_of_line = B_TRUE;
 615         static boolean_t local_echo = B_FALSE;
 616         char ibuf[ZLOGIN_BUFSIZ];
 617         int nbytes;
 618         char *buf = ibuf;
 619         char c = *buf;
 620 
 621         nbytes = read(STDIN_FILENO, ibuf, ZLOGIN_RDBUFSIZ);
 622         if (nbytes == -1 && (errno != EINTR || dead))
 623                 return (-1);
 624 
 625         if (nbytes == -1)       /* The read was interrupted. */
 626                 return (0);
 627 
 628         /* 0 read means EOF, close the pipe to the child */
 629         if (nbytes == 0)
 630                 return (1);
 631 
 632         for (c = *buf; nbytes > 0; c = *buf, --nbytes) {
 633                 buf++;
 634                 if (beginning_of_line && !nocmdchar) {
 635                         beginning_of_line = B_FALSE;
 636                         if (c == cmdchar) {
 637                                 local_echo = B_TRUE;
 638                                 continue;
 639                         }
 640                 } else if (local_echo) {
 641                         local_echo = B_FALSE;
 642                         if (c == '.' || c == effective_termios.c_cc[VEOF]) {
 643                                 char cc[CANONIFY_LEN];
 644 
 645                                 canonify(c, cc);
 646                                 (void) write(STDOUT_FILENO, &cmdchar, 1);
 647                                 (void) write(STDOUT_FILENO, cc, strlen(cc));
 648                                 return (-1);
 649                         }
 650                 }
 651 retry:
 652                 if (write(outfd, &c, 1) <= 0) {
 653                         /*
 654                          * Since the fd we are writing to is opened with
 655                          * O_NONBLOCK it is possible to get EAGAIN if the
 656                          * pipe is full.  One way this could happen is if we
 657                          * are writing a lot of data into the pipe in this loop
 658                          * and the application on the other end is echoing that
 659                          * data back out to its stdout.  The output pipe can
 660                          * fill up since we are stuck here in this loop and not
 661                          * draining the other pipe.  We can try to read some of
 662                          * the data to see if we can drain the pipe so that the
 663                          * application can continue to make progress.  The read
 664                          * is non-blocking so we won't hang here.  We also wait
 665                          * a bit before retrying since there could be other
 666                          * reasons why the pipe is full and we don't want to
 667                          * continuously retry.
 668                          */
 669                         if (errno == EAGAIN) {
 670                                 struct timespec rqtp;
 671                                 int ln;
 672                                 char obuf[ZLOGIN_BUFSIZ];
 673 
 674                                 if ((ln = read(infd, obuf, ZLOGIN_BUFSIZ)) > 0)
 675                                         (void) write(STDOUT_FILENO, obuf, ln);
 676 
 677                                 /* sleep for 10 milliseconds */
 678                                 rqtp.tv_sec = 0;
 679                                 rqtp.tv_nsec = 10 * (NANOSEC / MILLISEC);
 680                                 (void) nanosleep(&rqtp, NULL);
 681                                 if (!dead)
 682                                         goto retry;
 683                         }
 684 
 685                         return (-1);
 686                 }
 687                 beginning_of_line = (c == '\r' || c == '\n' ||
 688                     c == effective_termios.c_cc[VKILL] ||
 689                     c == effective_termios.c_cc[VEOL] ||
 690                     c == effective_termios.c_cc[VSUSP] ||
 691                     c == effective_termios.c_cc[VINTR]);
 692         }
 693         return (0);
 694 }
 695 
 696 /*
 697  * This function prevents deadlock between zlogin and the application in the
 698  * zone that it is talking to.  This can happen when we read from zlogin's
 699  * stdin and write the data down the pipe to the application.  If the pipe
 700  * is full, we'll block in the write.  Because zlogin could be blocked in
 701  * the write, it would never read the application's stdout/stderr so the
 702  * application can then block on those writes (when the pipe fills up).  If the
 703  * the application gets blocked this way, it can never get around to reading
 704  * its stdin so that zlogin can unblock from its write.  Once in this state,
 705  * the two processes are deadlocked.
 706  *
 707  * To prevent this, we want to verify that we can write into the pipe before we
 708  * read from our stdin.  If the pipe already is pretty full, we bypass the read
 709  * for now.  We'll circle back here again after the poll() so that we can
 710  * try again.  When this function is called, we already know there is data
 711  * ready to read on STDIN_FILENO.  We return -1 if there is a problem, 1 if
 712  * stdin is EOF, and 0 if everything is ok (even though we might not have
 713  * read/written any data into the pipe on this iteration).
 714  */
 715 static int
 716 process_raw_input(int stdin_fd, int appin_fd)
 717 {
 718         int cc;
 719         struct stat64 sb;
 720         char ibuf[ZLOGIN_RDBUFSIZ];
 721 
 722         /* Check how much data is already in the pipe */
 723         if (fstat64(appin_fd, &sb) == -1) {
 724                 perror("stat failed");
 725                 return (-1);
 726         }
 727 
 728         if (dead)
 729                 return (-1);
 730 
 731         /*
 732          * The pipe already has a lot of data in it,  don't write any more
 733          * right now.
 734          */
 735         if (sb.st_size >= HI_WATER)
 736                 return (0);
 737 
 738         cc = read(STDIN_FILENO, ibuf, ZLOGIN_RDBUFSIZ);
 739         if (cc == -1 && (errno != EINTR || dead))
 740                 return (-1);
 741 
 742         if (cc == -1)   /* The read was interrupted. */
 743                 return (0);
 744 
 745         /* 0 read means EOF, close the pipe to the child */
 746         if (cc == 0)
 747                 return (1);
 748 
 749         /*
 750          * stdin_fd is stdin of the target; so, the thing we'll write the user
 751          * data *to*.
 752          */
 753         if (write(stdin_fd, ibuf, cc) == -1)
 754                 return (-1);
 755 
 756         return (0);
 757 }
 758 
 759 /*
 760  * Write the output from the application running in the zone.  We can get
 761  * a signal during the write (usually it would be SIGCHLD when the application
 762  * has exited) so we loop to make sure we have written all of the data we read.
 763  */
 764 static int
 765 process_output(int in_fd, int out_fd)
 766 {
 767         int wrote = 0;
 768         int cc;
 769         char ibuf[ZLOGIN_BUFSIZ];
 770 
 771         cc = read(in_fd, ibuf, ZLOGIN_BUFSIZ);
 772         if (cc == -1 && (errno != EINTR || dead))
 773                 return (-1);
 774         if (cc == 0)    /* EOF */
 775                 return (-1);
 776         if (cc == -1)   /* The read was interrupted. */
 777                 return (0);
 778 
 779         do {
 780                 int len;
 781 
 782                 len = write(out_fd, ibuf + wrote, cc - wrote);
 783                 if (len == -1 && errno != EINTR)
 784                         return (-1);
 785                 if (len != -1)
 786                         wrote += len;
 787         } while (wrote < cc);
 788 
 789         return (0);
 790 }
 791 
 792 /*
 793  * This is the main I/O loop, and is shared across all zlogin modes.
 794  * Parameters:
 795  *      stdin_fd:  The fd representing 'stdin' for the slave side; input to
 796  *                 the zone will be written here.
 797  *
 798  *      appin_fd:  The fd representing the other end of the 'stdin' pipe (when
 799  *                 we're running non-interactive); used in process_raw_input
 800  *                 to ensure we don't fill up the application's stdin pipe.
 801  *
 802  *      stdout_fd: The fd representing 'stdout' for the slave side; output
 803  *                 from the zone will arrive here.
 804  *
 805  *      stderr_fd: The fd representing 'stderr' for the slave side; output
 806  *                 from the zone will arrive here.
 807  *
 808  *      raw_mode:  If TRUE, then no processing (for example, for '~.') will
 809  *                 be performed on the input coming from STDIN.
 810  *
 811  * stderr_fd may be specified as -1 if there is no stderr (only non-interactive
 812  * mode supplies a stderr).
 813  *
 814  */
 815 static void
 816 doio(int stdin_fd, int appin_fd, int stdout_fd, int stderr_fd, int sig_fd,
 817     boolean_t raw_mode)
 818 {
 819         struct pollfd pollfds[4];
 820         char ibuf[ZLOGIN_BUFSIZ];
 821         int cc, ret;
 822 
 823         /* read from stdout of zone and write to stdout of global zone */
 824         pollfds[0].fd = stdout_fd;
 825         pollfds[0].events = POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI;
 826 
 827         /* read from stderr of zone and write to stderr of global zone */
 828         pollfds[1].fd = stderr_fd;
 829         pollfds[1].events = pollfds[0].events;
 830 
 831         /* read from stdin of global zone and write to stdin of zone */
 832         pollfds[2].fd = STDIN_FILENO;
 833         pollfds[2].events = pollfds[0].events;
 834 
 835         /* read from signalling pipe so we know when child dies */
 836         pollfds[3].fd = sig_fd;
 837         pollfds[3].events = pollfds[0].events;
 838 
 839         for (;;) {
 840                 pollfds[0].revents = pollfds[1].revents =
 841                     pollfds[2].revents = pollfds[3].revents = 0;
 842 
 843                 if (dead)
 844                         break;
 845 
 846                 /*
 847                  * There is a race condition here where we can receive the
 848                  * child death signal, set the dead flag, but since we have
 849                  * passed the test above, we would go into poll and hang.
 850                  * To avoid this we use the sig_fd as an additional poll fd.
 851                  * The signal handler writes into the other end of this pipe
 852                  * when the child dies so that the poll will always see that
 853                  * input and proceed.  We just loop around at that point and
 854                  * then notice the dead flag.
 855                  */
 856 
 857                 ret = poll(pollfds,
 858                     sizeof (pollfds) / sizeof (struct pollfd), -1);
 859 
 860                 if (ret == -1 && errno != EINTR) {
 861                         perror("poll failed");
 862                         break;
 863                 }
 864 
 865                 if (errno == EINTR && dead) {
 866                         break;
 867                 }
 868 
 869                 /* event from master side stdout */
 870                 if (pollfds[0].revents) {
 871                         if (pollfds[0].revents &
 872                             (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
 873                                 if (process_output(stdout_fd, STDOUT_FILENO)
 874                                     != 0)
 875                                         break;
 876                         } else {
 877                                 pollerr = pollfds[0].revents;
 878                                 break;
 879                         }
 880                 }
 881 
 882                 /* event from master side stderr */
 883                 if (pollfds[1].revents) {
 884                         if (pollfds[1].revents &
 885                             (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
 886                                 if (process_output(stderr_fd, STDERR_FILENO)
 887                                     != 0)
 888                                         break;
 889                         } else {
 890                                 pollerr = pollfds[1].revents;
 891                                 break;
 892                         }
 893                 }
 894 
 895                 /* event from user STDIN side */
 896                 if (pollfds[2].revents) {
 897                         if (pollfds[2].revents &
 898                             (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
 899                                 /*
 900                                  * stdin fd is stdin of the target; so,
 901                                  * the thing we'll write the user data *to*.
 902                                  *
 903                                  * Also, unlike on the output side, we
 904                                  * close the pipe on a zero-length message.
 905                                  */
 906                                 int res;
 907 
 908                                 if (raw_mode)
 909                                         res = process_raw_input(stdin_fd,
 910                                             appin_fd);
 911                                 else
 912                                         res = process_user_input(stdin_fd,
 913                                             stdout_fd);
 914 
 915                                 if (res < 0)
 916                                         break;
 917                                 if (res > 0) {
 918                                         /* EOF (close) child's stdin_fd */
 919                                         pollfds[2].fd = -1;
 920                                         while ((res = close(stdin_fd)) != 0 &&
 921                                             errno == EINTR)
 922                                                 ;
 923                                         if (res != 0)
 924                                                 break;
 925                                 }
 926 
 927                         } else if (raw_mode && pollfds[2].revents & POLLHUP) {
 928                                 /*
 929                                  * It's OK to get a POLLHUP on STDIN-- it
 930                                  * always happens if you do:
 931                                  *
 932                                  * echo foo | zlogin <zone> <command>
 933                                  *
 934                                  * We reset fd to -1 in this case to clear
 935                                  * the condition and close the pipe (EOF) to
 936                                  * the other side in order to wrap things up.
 937                                  */
 938                                 int res;
 939 
 940                                 pollfds[2].fd = -1;
 941                                 while ((res = close(stdin_fd)) != 0 &&
 942                                     errno == EINTR)
 943                                         ;
 944                                 if (res != 0)
 945                                         break;
 946                         } else {
 947                                 pollerr = pollfds[2].revents;
 948                                 break;
 949                         }
 950                 }
 951         }
 952 
 953         /*
 954          * We are in the midst of dying, but try to poll with a short
 955          * timeout to see if we can catch the last bit of I/O from the
 956          * children.
 957          */
 958 retry:
 959         pollfds[0].revents = pollfds[1].revents = 0;
 960         (void) poll(pollfds, 2, 100);
 961         if (pollfds[0].revents &
 962             (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
 963                 if ((cc = read(stdout_fd, ibuf, ZLOGIN_BUFSIZ)) > 0) {
 964                         (void) write(STDOUT_FILENO, ibuf, cc);
 965                         goto retry;
 966                 }
 967         }
 968         if (pollfds[1].revents &
 969             (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) {
 970                 if ((cc = read(stderr_fd, ibuf, ZLOGIN_BUFSIZ)) > 0) {
 971                         (void) write(STDERR_FILENO, ibuf, cc);
 972                         goto retry;
 973                 }
 974         }
 975 }
 976 
 977 /*
 978  * Fetch the user_cmd brand hook for getting a user's passwd(4) entry.
 979  */
 980 static const char *
 981 zone_get_user_cmd(brand_handle_t bh, const char *login, char *user_cmd,
 982     size_t len)
 983 {
 984         bzero(user_cmd, sizeof (user_cmd));
 985         if (brand_get_user_cmd(bh, login, user_cmd, len) != 0)
 986                 return (NULL);
 987 
 988         return (user_cmd);
 989 }
 990 
 991 /* From libc */
 992 extern int str2passwd(const char *, int, void *, char *, int);
 993 
 994 /*
 995  * exec() the user_cmd brand hook, and convert the output string to a
 996  * struct passwd.  This is to be called after zone_enter().
 997  *
 998  */
 999 static struct passwd *
1000 zone_get_user_pw(const char *user_cmd, struct passwd *pwent, char *pwbuf,
1001     int pwbuflen)
1002 {
1003         char pwline[NSS_BUFLEN_PASSWD];
1004         char *cin = NULL;
1005         FILE *fin;
1006         int status;
1007 
1008         assert(getzoneid() != GLOBAL_ZONEID);
1009 
1010         if ((fin = popen(user_cmd, "r")) == NULL)
1011                 return (NULL);
1012 
1013         while (cin == NULL && !feof(fin))
1014                 cin = fgets(pwline, sizeof (pwline), fin);
1015 
1016         if (cin == NULL) {
1017                 (void) pclose(fin);
1018                 return (NULL);
1019         }
1020 
1021         status = pclose(fin);
1022         if (!WIFEXITED(status))
1023                 return (NULL);
1024         if (WEXITSTATUS(status) != 0)
1025                 return (NULL);
1026 
1027         if (str2passwd(pwline, sizeof (pwline), pwent, pwbuf, pwbuflen) == 0)
1028                 return (pwent);
1029         else
1030                 return (NULL);
1031 }
1032 
1033 static char **
1034 zone_login_cmd(brand_handle_t bh, const char *login)
1035 {
1036         static char result_buf[ARG_MAX];
1037         char **new_argv, *ptr, *lasts;
1038         int n, a;
1039 
1040         /* Get the login command for the target zone. */
1041         bzero(result_buf, sizeof (result_buf));
1042 
1043         if (forced_login) {
1044                 if (brand_get_forcedlogin_cmd(bh, login,
1045                     result_buf, sizeof (result_buf)) != 0)
1046                         return (NULL);
1047         } else {
1048                 if (brand_get_login_cmd(bh, login,
1049                     result_buf, sizeof (result_buf)) != 0)
1050                         return (NULL);
1051         }
1052 
1053         /*
1054          * We got back a string that we'd like to execute.  But since
1055          * we're not doing the execution via a shell we'll need to convert
1056          * the exec string to an array of strings.  We'll do that here
1057          * but we're going to be very simplistic about it and break stuff
1058          * up based on spaces.  We're not even going to support any kind
1059          * of quoting or escape characters.  It's truly amazing that
1060          * there is no library function in OpenSolaris to do this for us.
1061          */
1062 
1063         /*
1064          * Be paranoid.  Since we're deliniating based on spaces make
1065          * sure there are no adjacent spaces.
1066          */
1067         if (strstr(result_buf, "  ") != NULL)
1068                 return (NULL);
1069 
1070         /* Remove any trailing whitespace.  */
1071         n = strlen(result_buf);
1072         if (result_buf[n - 1] == ' ')
1073                 result_buf[n - 1] = '\0';
1074 
1075         /* Count how many elements there are in the exec string. */
1076         ptr = result_buf;
1077         for (n = 2; ((ptr = strchr(ptr + 1, (int)' ')) != NULL); n++)
1078                 ;
1079 
1080         /* Allocate the argv array that we're going to return. */
1081         if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1082                 return (NULL);
1083 
1084         /* Tokenize the exec string and return. */
1085         a = 0;
1086         new_argv[a++] = result_buf;
1087         if (n > 2) {
1088                 (void) strtok_r(result_buf, " ", &lasts);
1089                 while ((new_argv[a++] = strtok_r(NULL, " ", &lasts)) != NULL)
1090                         ;
1091         } else {
1092                 new_argv[a++] = NULL;
1093         }
1094         assert(n == a);
1095         return (new_argv);
1096 }
1097 
1098 /*
1099  * Prepare argv array for exec'd process; if we're passing commands to the
1100  * new process, then use su(1M) to do the invocation.  Otherwise, use
1101  * 'login -z <from_zonename> -f' (-z is an undocumented option which tells
1102  * login that we're coming from another zone, and to disregard its CONSOLE
1103  * checks).
1104  */
1105 static char **
1106 prep_args(brand_handle_t bh, const char *login, char **argv)
1107 {
1108         int argc = 0, a = 0, i, n = -1;
1109         char **new_argv;
1110 
1111         if (argv != NULL) {
1112                 size_t subshell_len = 1;
1113                 char *subshell;
1114 
1115                 while (argv[argc] != NULL)
1116                         argc++;
1117 
1118                 for (i = 0; i < argc; i++) {
1119                         subshell_len += strlen(argv[i]) + 1;
1120                 }
1121                 if ((subshell = calloc(1, subshell_len)) == NULL)
1122                         return (NULL);
1123 
1124                 for (i = 0; i < argc; i++) {
1125                         (void) strcat(subshell, argv[i]);
1126                         (void) strcat(subshell, " ");
1127                 }
1128 
1129                 if (failsafe) {
1130                         n = 4;
1131                         if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1132                                 return (NULL);
1133 
1134                         new_argv[a++] = FAILSAFESHELL;
1135                 } else {
1136                         n = 5;
1137                         if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1138                                 return (NULL);
1139 
1140                         new_argv[a++] = SUPATH;
1141                         if (strcmp(login, "root") != 0) {
1142                                 new_argv[a++] = "-";
1143                                 n++;
1144                         }
1145                         new_argv[a++] = (char *)login;
1146                 }
1147                 new_argv[a++] = "-c";
1148                 new_argv[a++] = subshell;
1149                 new_argv[a++] = NULL;
1150                 assert(a == n);
1151         } else {
1152                 if (failsafe) {
1153                         n = 2;
1154                         if ((new_argv = malloc(sizeof (char *) * n)) == NULL)
1155                                 return (NULL);
1156                         new_argv[a++] = FAILSAFESHELL;
1157                         new_argv[a++] = NULL;
1158                         assert(n == a);
1159                 } else {
1160                         new_argv = zone_login_cmd(bh, login);
1161                 }
1162         }
1163 
1164         return (new_argv);
1165 }
1166 
1167 /*
1168  * Helper routine for prep_env below.
1169  */
1170 static char *
1171 add_env(char *name, char *value)
1172 {
1173         size_t sz = strlen(name) + strlen(value) + 2; /* name, =, value, NUL */
1174         char *str;
1175 
1176         if ((str = malloc(sz)) == NULL)
1177                 return (NULL);
1178 
1179         (void) snprintf(str, sz, "%s=%s", name, value);
1180         return (str);
1181 }
1182 
1183 /*
1184  * Prepare envp array for exec'd process.
1185  */
1186 static char **
1187 prep_env()
1188 {
1189         int e = 0, size = 1;
1190         char **new_env, *estr;
1191         char *term = getenv("TERM");
1192 
1193         size++; /* for $PATH */
1194         if (term != NULL)
1195                 size++;
1196 
1197         /*
1198          * In failsafe mode we set $HOME, since '-l' isn't valid in this mode.
1199          * We also set $SHELL, since neither login nor su will be around to do
1200          * it.
1201          */
1202         if (failsafe)
1203                 size += 2;
1204 
1205         if ((new_env = malloc(sizeof (char *) * size)) == NULL)
1206                 return (NULL);
1207 
1208         if ((estr = add_env("PATH", DEF_PATH)) == NULL)
1209                 return (NULL);
1210         new_env[e++] = estr;
1211 
1212         if (term != NULL) {
1213                 if ((estr = add_env("TERM", term)) == NULL)
1214                         return (NULL);
1215                 new_env[e++] = estr;
1216         }
1217 
1218         if (failsafe) {
1219                 if ((estr = add_env("HOME", "/")) == NULL)
1220                         return (NULL);
1221                 new_env[e++] = estr;
1222 
1223                 if ((estr = add_env("SHELL", FAILSAFESHELL)) == NULL)
1224                         return (NULL);
1225                 new_env[e++] = estr;
1226         }
1227 
1228         new_env[e++] = NULL;
1229 
1230         assert(e == size);
1231 
1232         return (new_env);
1233 }
1234 
1235 /*
1236  * Finish the preparation of the envp array for exec'd non-interactive
1237  * zlogins.  This is called in the child process *after* we zone_enter(), since
1238  * it derives things we can only know within the zone, such as $HOME, $SHELL,
1239  * etc.  We need only do this in the non-interactive, mode, since otherwise
1240  * login(1) will do it.  We don't do this in failsafe mode, since it presents
1241  * additional ways in which the command could fail, and we'd prefer to avoid
1242  * that.
1243  */
1244 static char **
1245 prep_env_noninteractive(const char *user_cmd, char **env)
1246 {
1247         size_t size;
1248         char **new_env;
1249         int e, i;
1250         char *estr;
1251         char varmail[_LOGNAME_MAX + 11]; /* strlen(/var/mail/) = 10, NUL */
1252         char pwbuf[NSS_BUFLEN_PASSWD + 1];
1253         struct passwd pwent;
1254         struct passwd *pw = NULL;
1255 
1256         assert(env != NULL);
1257         assert(failsafe == 0);
1258 
1259         /*
1260          * Exec the "user_cmd" brand hook to get a pwent for the
1261          * login user.  If this fails, HOME will be set to "/", SHELL
1262          * will be set to $DEFAULTSHELL, and we will continue to exec
1263          * SUPATH <login> -c <cmd>.
1264          */
1265         pw = zone_get_user_pw(user_cmd, &pwent, pwbuf, sizeof (pwbuf));
1266 
1267         /*
1268          * Get existing envp size.
1269          */
1270         for (size = 0; env[size] != NULL; size++)
1271                 ;
1272 
1273         e = size;
1274 
1275         /*
1276          * Finish filling out the environment; we duplicate the environment
1277          * setup described in login(1), for lack of a better precedent.
1278          */
1279         if (pw != NULL)
1280                 size += 3;      /* LOGNAME, HOME, MAIL */
1281         else
1282                 size += 1;      /* HOME */
1283 
1284         size++; /* always fill in SHELL */
1285         size++; /* terminating NULL */
1286 
1287         if ((new_env = malloc(sizeof (char *) * size)) == NULL)
1288                 goto malloc_fail;
1289 
1290         /*
1291          * Copy existing elements of env into new_env.
1292          */
1293         for (i = 0; env[i] != NULL; i++) {
1294                 if ((new_env[i] = strdup(env[i])) == NULL)
1295                         goto malloc_fail;
1296         }
1297         assert(e == i);
1298 
1299         if (pw != NULL) {
1300                 if ((estr = add_env("LOGNAME", pw->pw_name)) == NULL)
1301                         goto malloc_fail;
1302                 new_env[e++] = estr;
1303 
1304                 if ((estr = add_env("HOME", pw->pw_dir)) == NULL)
1305                         goto malloc_fail;
1306                 new_env[e++] = estr;
1307 
1308                 if (chdir(pw->pw_dir) != 0)
1309                         zerror(gettext("Could not chdir to home directory "
1310                             "%s: %s"), pw->pw_dir, strerror(errno));
1311 
1312                 (void) snprintf(varmail, sizeof (varmail), "/var/mail/%s",
1313                     pw->pw_name);
1314                 if ((estr = add_env("MAIL", varmail)) == NULL)
1315                         goto malloc_fail;
1316                 new_env[e++] = estr;
1317         } else {
1318                 if ((estr = add_env("HOME", "/")) == NULL)
1319                         goto malloc_fail;
1320                 new_env[e++] = estr;
1321         }
1322 
1323         if (pw != NULL && strlen(pw->pw_shell) > 0) {
1324                 if ((estr = add_env("SHELL", pw->pw_shell)) == NULL)
1325                         goto malloc_fail;
1326                 new_env[e++] = estr;
1327         } else {
1328                 if ((estr = add_env("SHELL", DEFAULTSHELL)) == NULL)
1329                         goto malloc_fail;
1330                 new_env[e++] = estr;
1331         }
1332 
1333         new_env[e++] = NULL;    /* add terminating NULL */
1334 
1335         assert(e == size);
1336         return (new_env);
1337 
1338 malloc_fail:
1339         zperror(gettext("failed to allocate memory for process environment"));
1340         return (NULL);
1341 }
1342 
1343 static int
1344 close_func(void *slavefd, int fd)
1345 {
1346         if (fd != *(int *)slavefd)
1347                 (void) close(fd);
1348         return (0);
1349 }
1350 
1351 static void
1352 set_cmdchar(char *cmdcharstr)
1353 {
1354         char c;
1355         long lc;
1356 
1357         if ((c = *cmdcharstr) != '\\') {
1358                 cmdchar = c;
1359                 return;
1360         }
1361 
1362         c = cmdcharstr[1];
1363         if (c == '\0' || c == '\\') {
1364                 cmdchar = '\\';
1365                 return;
1366         }
1367 
1368         if (c < '0' || c > '7') {
1369                 zerror(gettext("Unrecognized escape character option %s"),
1370                     cmdcharstr);
1371                 usage();
1372         }
1373 
1374         lc = strtol(cmdcharstr + 1, NULL, 8);
1375         if (lc < 0 || lc > 255) {
1376                 zerror(gettext("Octal escape character '%s' too large"),
1377                     cmdcharstr);
1378                 usage();
1379         }
1380         cmdchar = (char)lc;
1381 }
1382 
1383 static int
1384 setup_utmpx(char *slavename)
1385 {
1386         struct utmpx ut;
1387 
1388         bzero(&ut, sizeof (ut));
1389         (void) strncpy(ut.ut_user, ".zlogin", sizeof (ut.ut_user));
1390         (void) strncpy(ut.ut_line, slavename, sizeof (ut.ut_line));
1391         ut.ut_pid = getpid();
1392         ut.ut_id[0] = 'z';
1393         ut.ut_id[1] = ut.ut_id[2] = ut.ut_id[3] = (char)SC_WILDC;
1394         ut.ut_type = LOGIN_PROCESS;
1395         (void) time(&ut.ut_tv.tv_sec);
1396 
1397         if (makeutx(&ut) == NULL) {
1398                 zerror(gettext("makeutx failed"));
1399                 return (-1);
1400         }
1401         return (0);
1402 }
1403 
1404 static void
1405 release_lock_file(int lockfd)
1406 {
1407         (void) close(lockfd);
1408 }
1409 
1410 static int
1411 grab_lock_file(const char *zone_name, int *lockfd)
1412 {
1413         char pathbuf[PATH_MAX];
1414         struct flock flock;
1415 
1416         if (mkdir(ZONES_TMPDIR, S_IRWXU) < 0 && errno != EEXIST) {
1417                 zerror(gettext("could not mkdir %s: %s"), ZONES_TMPDIR,
1418                     strerror(errno));
1419                 return (-1);
1420         }
1421         (void) chmod(ZONES_TMPDIR, S_IRWXU);
1422         (void) snprintf(pathbuf, sizeof (pathbuf), "%s/%s.zoneadm.lock",
1423             ZONES_TMPDIR, zone_name);
1424 
1425         if ((*lockfd = open(pathbuf, O_RDWR|O_CREAT, S_IRUSR|S_IWUSR)) < 0) {
1426                 zerror(gettext("could not open %s: %s"), pathbuf,
1427                     strerror(errno));
1428                 return (-1);
1429         }
1430         /*
1431          * Lock the file to synchronize with other zoneadmds
1432          */
1433         flock.l_type = F_WRLCK;
1434         flock.l_whence = SEEK_SET;
1435         flock.l_start = (off_t)0;
1436         flock.l_len = (off_t)0;
1437         if (fcntl(*lockfd, F_SETLKW, &flock) < 0) {
1438                 zerror(gettext("unable to lock %s: %s"), pathbuf,
1439                     strerror(errno));
1440                 release_lock_file(*lockfd);
1441                 return (-1);
1442         }
1443         return (Z_OK);
1444 }
1445 
1446 static int
1447 start_zoneadmd(const char *zone_name)
1448 {
1449         pid_t retval;
1450         int pstatus = 0, error = -1, lockfd, doorfd;
1451         struct door_info info;
1452         char doorpath[MAXPATHLEN];
1453 
1454         (void) snprintf(doorpath, sizeof (doorpath), ZONE_DOOR_PATH, zone_name);
1455 
1456         if (grab_lock_file(zone_name, &lockfd) != Z_OK)
1457                 return (-1);
1458         /*
1459          * We must do the door check with the lock held.  Otherwise, we
1460          * might race against another zoneadm/zlogin process and wind
1461          * up with two processes trying to start zoneadmd at the same
1462          * time.  zoneadmd will detect this, and fail, but we prefer this
1463          * to be as seamless as is practical, from a user perspective.
1464          */
1465         if ((doorfd = open(doorpath, O_RDONLY)) < 0) {
1466                 if (errno != ENOENT) {
1467                         zerror("failed to open %s: %s", doorpath,
1468                             strerror(errno));
1469                         goto out;
1470                 }
1471         } else {
1472                 /*
1473                  * Seems to be working ok.
1474                  */
1475                 if (door_info(doorfd, &info) == 0 &&
1476                     ((info.di_attributes & DOOR_REVOKED) == 0)) {
1477                         error = 0;
1478                         goto out;
1479                 }
1480         }
1481 
1482         if ((child_pid = fork()) == -1) {
1483                 zperror(gettext("could not fork"));
1484                 goto out;
1485         } else if (child_pid == 0) {
1486                 /* child process */
1487                 (void) execl("/usr/lib/zones/zoneadmd", "zoneadmd", "-z",
1488                     zone_name, NULL);
1489                 zperror(gettext("could not exec zoneadmd"));
1490                 _exit(1);
1491         }
1492 
1493         /* parent process */
1494         do {
1495                 retval = waitpid(child_pid, &pstatus, 0);
1496         } while (retval != child_pid);
1497         if (WIFSIGNALED(pstatus) ||
1498             (WIFEXITED(pstatus) && WEXITSTATUS(pstatus) != 0)) {
1499                 zerror(gettext("could not start %s"), "zoneadmd");
1500                 goto out;
1501         }
1502         error = 0;
1503 out:
1504         release_lock_file(lockfd);
1505         (void) close(doorfd);
1506         return (error);
1507 }
1508 
1509 static int
1510 init_template(void)
1511 {
1512         int fd;
1513         int err = 0;
1514 
1515         fd = open64(CTFS_ROOT "/process/template", O_RDWR);
1516         if (fd == -1)
1517                 return (-1);
1518 
1519         /*
1520          * zlogin doesn't do anything with the contract.
1521          * Deliver no events, don't inherit, and allow it to be orphaned.
1522          */
1523         err |= ct_tmpl_set_critical(fd, 0);
1524         err |= ct_tmpl_set_informative(fd, 0);
1525         err |= ct_pr_tmpl_set_fatal(fd, CT_PR_EV_HWERR);
1526         err |= ct_pr_tmpl_set_param(fd, CT_PR_PGRPONLY | CT_PR_REGENT);
1527         if (err || ct_tmpl_activate(fd)) {
1528                 (void) close(fd);
1529                 return (-1);
1530         }
1531 
1532         return (fd);
1533 }
1534 
1535 static int
1536 noninteractive_login(char *zonename, const char *user_cmd, zoneid_t zoneid,
1537     char **new_args, char **new_env)
1538 {
1539         pid_t retval;
1540         int stdin_pipe[2], stdout_pipe[2], stderr_pipe[2], dead_child_pipe[2];
1541         int child_status;
1542         int tmpl_fd;
1543         sigset_t block_cld;
1544 
1545         if ((tmpl_fd = init_template()) == -1) {
1546                 reset_tty();
1547                 zperror(gettext("could not create contract"));
1548                 return (1);
1549         }
1550 
1551         if (pipe(stdin_pipe) != 0) {
1552                 zperror(gettext("could not create STDIN pipe"));
1553                 return (1);
1554         }
1555         /*
1556          * When the user types ^D, we get a zero length message on STDIN.
1557          * We need to echo that down the pipe to send it to the other side;
1558          * but by default, pipes don't propagate zero-length messages.  We
1559          * toggle that behavior off using I_SWROPT.  See streamio(7i).
1560          */
1561         if (ioctl(stdin_pipe[0], I_SWROPT, SNDZERO) != 0) {
1562                 zperror(gettext("could not configure STDIN pipe"));
1563                 return (1);
1564 
1565         }
1566         if (pipe(stdout_pipe) != 0) {
1567                 zperror(gettext("could not create STDOUT pipe"));
1568                 return (1);
1569         }
1570         if (pipe(stderr_pipe) != 0) {
1571                 zperror(gettext("could not create STDERR pipe"));
1572                 return (1);
1573         }
1574 
1575         if (pipe(dead_child_pipe) != 0) {
1576                 zperror(gettext("could not create signalling pipe"));
1577                 return (1);
1578         }
1579         close_on_sig = dead_child_pipe[0];
1580 
1581         /*
1582          * If any of the pipe FD's winds up being less than STDERR, then we
1583          * have a mess on our hands-- and we are lacking some of the I/O
1584          * streams we would expect anyway.  So we bail.
1585          */
1586         if (stdin_pipe[0] <= STDERR_FILENO ||
1587             stdin_pipe[1] <= STDERR_FILENO ||
1588             stdout_pipe[0] <= STDERR_FILENO ||
1589             stdout_pipe[1] <= STDERR_FILENO ||
1590             stderr_pipe[0] <= STDERR_FILENO ||
1591             stderr_pipe[1] <= STDERR_FILENO ||
1592             dead_child_pipe[0] <= STDERR_FILENO ||
1593             dead_child_pipe[1] <= STDERR_FILENO) {
1594                 zperror(gettext("process lacks valid STDIN, STDOUT, STDERR"));
1595                 return (1);
1596         }
1597 
1598         if (prefork_dropprivs() != 0) {
1599                 zperror(gettext("could not allocate privilege set"));
1600                 return (1);
1601         }
1602 
1603         (void) sigset(SIGCLD, sigcld);
1604         (void) sigemptyset(&block_cld);
1605         (void) sigaddset(&block_cld, SIGCLD);
1606         (void) sigprocmask(SIG_BLOCK, &block_cld, NULL);
1607 
1608         if ((child_pid = fork()) == -1) {
1609                 (void) ct_tmpl_clear(tmpl_fd);
1610                 (void) close(tmpl_fd);
1611                 zperror(gettext("could not fork"));
1612                 return (1);
1613         } else if (child_pid == 0) { /* child process */
1614                 (void) ct_tmpl_clear(tmpl_fd);
1615 
1616                 /*
1617                  * Do a dance to get the pipes hooked up as FD's 0, 1 and 2.
1618                  */
1619                 (void) close(STDIN_FILENO);
1620                 (void) close(STDOUT_FILENO);
1621                 (void) close(STDERR_FILENO);
1622                 (void) dup2(stdin_pipe[1], STDIN_FILENO);
1623                 (void) dup2(stdout_pipe[1], STDOUT_FILENO);
1624                 (void) dup2(stderr_pipe[1], STDERR_FILENO);
1625                 (void) closefrom(STDERR_FILENO + 1);
1626 
1627                 (void) sigset(SIGCLD, SIG_DFL);
1628                 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
1629                 /*
1630                  * In case any of stdin, stdout or stderr are streams,
1631                  * anchor them to prevent malicious I_POPs.
1632                  */
1633                 (void) ioctl(STDIN_FILENO, I_ANCHOR);
1634                 (void) ioctl(STDOUT_FILENO, I_ANCHOR);
1635                 (void) ioctl(STDERR_FILENO, I_ANCHOR);
1636 
1637                 if (zone_enter(zoneid) == -1) {
1638                         zerror(gettext("could not enter zone %s: %s"),
1639                             zonename, strerror(errno));
1640                         _exit(1);
1641                 }
1642 
1643                 /*
1644                  * For non-native zones, tell libc where it can find locale
1645                  * specific getttext() messages.
1646                  */
1647                 if (access("/.SUNWnative/usr/lib/locale", R_OK) == 0)
1648                         (void) bindtextdomain(TEXT_DOMAIN,
1649                             "/.SUNWnative/usr/lib/locale");
1650                 else if (access("/native/usr/lib/locale", R_OK) == 0)
1651                         (void) bindtextdomain(TEXT_DOMAIN,
1652                             "/native/usr/lib/locale");
1653 
1654                 if (!failsafe)
1655                         new_env = prep_env_noninteractive(user_cmd, new_env);
1656 
1657                 if (new_env == NULL) {
1658                         _exit(1);
1659                 }
1660 
1661                 /*
1662                  * Move into a new process group; the zone_enter will have
1663                  * placed us into zsched's session, and we want to be in
1664                  * a unique process group.
1665                  */
1666                 (void) setpgid(getpid(), getpid());
1667 
1668                 /*
1669                  * The child needs to run as root to
1670                  * execute the su program.
1671                  */
1672                 if (setuid(0) == -1) {
1673                         zperror(gettext("insufficient privilege"));
1674                         return (1);
1675                 }
1676 
1677                 (void) execve(new_args[0], new_args, new_env);
1678                 zperror(gettext("exec failure"));
1679                 _exit(1);
1680         }
1681         /* parent */
1682 
1683         /* close pipe sides written by child */
1684         (void) close(stdout_pipe[1]);
1685         (void) close(stderr_pipe[1]);
1686 
1687         (void) sigset(SIGINT, sig_forward);
1688 
1689         postfork_dropprivs();
1690 
1691         (void) ct_tmpl_clear(tmpl_fd);
1692         (void) close(tmpl_fd);
1693 
1694         (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
1695         doio(stdin_pipe[0], stdin_pipe[1], stdout_pipe[0], stderr_pipe[0],
1696             dead_child_pipe[1], B_TRUE);
1697         do {
1698                 retval = waitpid(child_pid, &child_status, 0);
1699                 if (retval == -1) {
1700                         child_status = 0;
1701                 }
1702         } while (retval != child_pid && errno != ECHILD);
1703 
1704         return (WEXITSTATUS(child_status));
1705 }
1706 
1707 static char *
1708 get_username()
1709 {
1710         uid_t   uid;
1711         struct passwd *nptr;
1712 
1713         /*
1714          * Authorizations are checked to restrict access based on the
1715          * requested operation and zone name, It is assumed that the
1716          * program is running with all privileges, but that the real
1717          * user ID is that of the user or role on whose behalf we are
1718          * operating. So we start by getting the username that will be
1719          * used for subsequent authorization checks.
1720          */
1721 
1722         uid = getuid();
1723         if ((nptr = getpwuid(uid)) == NULL) {
1724                 zerror(gettext("could not get user name."));
1725                 _exit(1);
1726         }
1727         return (nptr->pw_name);
1728 }
1729 
1730 int
1731 main(int argc, char **argv)
1732 {
1733         int arg, console = 0;
1734         zoneid_t zoneid;
1735         zone_state_t st;
1736         char *login = "root";
1737         int lflag = 0;
1738         char *zonename = NULL;
1739         char **proc_args = NULL;
1740         char **new_args, **new_env;
1741         sigset_t block_cld;
1742         char devroot[MAXPATHLEN];
1743         char *slavename, slaveshortname[MAXPATHLEN];
1744         priv_set_t *privset;
1745         int tmpl_fd;
1746         char zonebrand[MAXNAMELEN];
1747         char default_brand[MAXNAMELEN];
1748         struct stat sb;
1749         char kernzone[ZONENAME_MAX];
1750         brand_handle_t bh;
1751         char user_cmd[MAXPATHLEN];
1752         char authname[MAXAUTHS];
1753 
1754         (void) setlocale(LC_ALL, "");
1755         (void) textdomain(TEXT_DOMAIN);
1756 
1757         (void) getpname(argv[0]);
1758         username = get_username();
1759 
1760         while ((arg = getopt(argc, argv, "ECR:Se:l:")) != EOF) {
1761                 switch (arg) {
1762                 case 'C':
1763                         console = 1;
1764                         break;
1765                 case 'E':
1766                         nocmdchar = 1;
1767                         break;
1768                 case 'R':       /* undocumented */
1769                         if (*optarg != '/') {
1770                                 zerror(gettext("root path must be absolute."));
1771                                 exit(2);
1772                         }
1773                         if (stat(optarg, &sb) == -1 || !S_ISDIR(sb.st_mode)) {
1774                                 zerror(
1775                                     gettext("root path must be a directory."));
1776                                 exit(2);
1777                         }
1778                         zonecfg_set_root(optarg);
1779                         break;
1780                 case 'S':
1781                         failsafe = 1;
1782                         break;
1783                 case 'e':
1784                         set_cmdchar(optarg);
1785                         break;
1786                 case 'l':
1787                         login = optarg;
1788                         lflag = 1;
1789                         break;
1790                 default:
1791                         usage();
1792                 }
1793         }
1794 
1795         if (console != 0 && lflag != 0) {
1796                 zerror(gettext("-l may not be specified for console login"));
1797                 usage();
1798         }
1799 
1800         if (console != 0 && failsafe != 0) {
1801                 zerror(gettext("-S may not be specified for console login"));
1802                 usage();
1803         }
1804 
1805         if (console != 0 && zonecfg_in_alt_root()) {
1806                 zerror(gettext("-R may not be specified for console login"));
1807                 exit(2);
1808         }
1809 
1810         if (failsafe != 0 && lflag != 0) {
1811                 zerror(gettext("-l may not be specified for failsafe login"));
1812                 usage();
1813         }
1814 
1815         if (optind == (argc - 1)) {
1816                 /*
1817                  * zone name, no process name; this should be an interactive
1818                  * as long as STDIN is really a tty.
1819                  */
1820                 if (isatty(STDIN_FILENO))
1821                         interactive = 1;
1822                 zonename = argv[optind];
1823         } else if (optind < (argc - 1)) {
1824                 if (console) {
1825                         zerror(gettext("Commands may not be specified for "
1826                             "console login."));
1827                         usage();
1828                 }
1829                 /* zone name and process name, and possibly some args */
1830                 zonename = argv[optind];
1831                 proc_args = &argv[optind + 1];
1832                 interactive = 0;
1833         } else {
1834                 usage();
1835         }
1836 
1837         if (getzoneid() != GLOBAL_ZONEID) {
1838                 zerror(gettext("'%s' may only be used from the global zone"),
1839                     pname);
1840                 return (1);
1841         }
1842 
1843         if (strcmp(zonename, GLOBAL_ZONENAME) == 0) {
1844                 zerror(gettext("'%s' not applicable to the global zone"),
1845                     pname);
1846                 return (1);
1847         }
1848 
1849         if (zone_get_state(zonename, &st) != Z_OK) {
1850                 zerror(gettext("zone '%s' unknown"), zonename);
1851                 return (1);
1852         }
1853 
1854         if (st < ZONE_STATE_INSTALLED) {
1855                 zerror(gettext("cannot login to a zone which is '%s'"),
1856                     zone_state_str(st));
1857                 return (1);
1858         }
1859 
1860         /*
1861          * In both console and non-console cases, we require all privs.
1862          * In the console case, because we may need to startup zoneadmd.
1863          * In the non-console case in order to do zone_enter(2), zonept()
1864          * and other tasks.
1865          */
1866 
1867         if ((privset = priv_allocset()) == NULL) {
1868                 zperror(gettext("priv_allocset failed"));
1869                 return (1);
1870         }
1871 
1872         if (getppriv(PRIV_EFFECTIVE, privset) != 0) {
1873                 zperror(gettext("getppriv failed"));
1874                 priv_freeset(privset);
1875                 return (1);
1876         }
1877 
1878         if (priv_isfullset(privset) == B_FALSE) {
1879                 zerror(gettext("You lack sufficient privilege to run "
1880                     "this command (all privs required)"));
1881                 priv_freeset(privset);
1882                 return (1);
1883         }
1884         priv_freeset(privset);
1885 
1886         /*
1887          * Check if user is authorized for requested usage of the zone
1888          */
1889 
1890         (void) snprintf(authname, MAXAUTHS, "%s%s%s",
1891             ZONE_MANAGE_AUTH, KV_OBJECT, zonename);
1892         if (chkauthattr(authname, username) == 0) {
1893                 if (console) {
1894                         zerror(gettext("%s is not authorized for console "
1895                             "access to  %s zone."),
1896                             username, zonename);
1897                         return (1);
1898                 } else {
1899                         (void) snprintf(authname, MAXAUTHS, "%s%s%s",
1900                             ZONE_LOGIN_AUTH, KV_OBJECT, zonename);
1901                         if (failsafe || !interactive) {
1902                                 zerror(gettext("%s is not authorized for  "
1903                                     "failsafe or non-interactive login "
1904                                     "to  %s zone."), username, zonename);
1905                                 return (1);
1906                         } else if (chkauthattr(authname, username) == 0) {
1907                                 zerror(gettext("%s is not authorized "
1908                                     " to login to %s zone."),
1909                                     username, zonename);
1910                                 return (1);
1911                         }
1912                 }
1913         } else {
1914                 forced_login = B_TRUE;
1915         }
1916 
1917         /*
1918          * The console is a separate case from the rest of the code; handle
1919          * it first.
1920          */
1921         if (console) {
1922                 /*
1923                  * Ensure that zoneadmd for this zone is running.
1924                  */
1925                 if (start_zoneadmd(zonename) == -1)
1926                         return (1);
1927 
1928                 /*
1929                  * Make contact with zoneadmd.
1930                  */
1931                 if (get_console_master(zonename) == -1)
1932                         return (1);
1933 
1934                 (void) printf(gettext("[Connected to zone '%s' console]\n"),
1935                     zonename);
1936 
1937                 if (set_tty_rawmode(STDIN_FILENO) == -1) {
1938                         reset_tty();
1939                         zperror(gettext("failed to set stdin pty to raw mode"));
1940                         return (1);
1941                 }
1942 
1943                 (void) sigset(SIGWINCH, sigwinch);
1944                 (void) sigwinch(0);
1945 
1946                 /*
1947                  * Run the I/O loop until we get disconnected.
1948                  */
1949                 doio(masterfd, -1, masterfd, -1, -1, B_FALSE);
1950                 reset_tty();
1951                 (void) printf(gettext("\n[Connection to zone '%s' console "
1952                     "closed]\n"), zonename);
1953 
1954                 return (0);
1955         }
1956 
1957         if (st != ZONE_STATE_RUNNING && st != ZONE_STATE_MOUNTED) {
1958                 zerror(gettext("login allowed only to running zones "
1959                     "(%s is '%s')."), zonename, zone_state_str(st));
1960                 return (1);
1961         }
1962 
1963         (void) strlcpy(kernzone, zonename, sizeof (kernzone));
1964         if (zonecfg_in_alt_root()) {
1965                 FILE *fp = zonecfg_open_scratch("", B_FALSE);
1966 
1967                 if (fp == NULL || zonecfg_find_scratch(fp, zonename,
1968                     zonecfg_get_root(), kernzone, sizeof (kernzone)) == -1) {
1969                         zerror(gettext("cannot find scratch zone %s"),
1970                             zonename);
1971                         if (fp != NULL)
1972                                 zonecfg_close_scratch(fp);
1973                         return (1);
1974                 }
1975                 zonecfg_close_scratch(fp);
1976         }
1977 
1978         if ((zoneid = getzoneidbyname(kernzone)) == -1) {
1979                 zerror(gettext("failed to get zoneid for zone '%s'"),
1980                     zonename);
1981                 return (1);
1982         }
1983 
1984         /*
1985          * We need the zone root path only if we are setting up a pty.
1986          */
1987         if (zone_get_devroot(zonename, devroot, sizeof (devroot)) == -1) {
1988                 zerror(gettext("could not get dev path for zone %s"),
1989                     zonename);
1990                 return (1);
1991         }
1992 
1993         if (zone_get_brand(zonename, zonebrand, sizeof (zonebrand)) != Z_OK) {
1994                 zerror(gettext("could not get brand for zone %s"), zonename);
1995                 return (1);
1996         }
1997         /*
1998          * In the alternate root environment, the only supported
1999          * operations are mount and unmount.  In this case, just treat
2000          * the zone as native if it is cluster.  Cluster zones can be
2001          * native for the purpose of LU or upgrade, and the cluster
2002          * brand may not exist in the miniroot (such as in net install
2003          * upgrade).
2004          */
2005         if (zonecfg_default_brand(default_brand,
2006             sizeof (default_brand)) != Z_OK) {
2007                 zerror(gettext("unable to determine default brand"));
2008                 return (1);
2009         }
2010         if (zonecfg_in_alt_root() &&
2011             strcmp(zonebrand, CLUSTER_BRAND_NAME) == 0) {
2012                 (void) strlcpy(zonebrand, default_brand, sizeof (zonebrand));
2013         }
2014 
2015         if ((bh = brand_open(zonebrand)) == NULL) {
2016                 zerror(gettext("could not open brand for zone %s"), zonename);
2017                 return (1);
2018         }
2019 
2020         if ((new_args = prep_args(bh, login, proc_args)) == NULL) {
2021                 zperror(gettext("could not assemble new arguments"));
2022                 brand_close(bh);
2023                 return (1);
2024         }
2025         /*
2026          * Get the brand specific user_cmd.  This command is used to get
2027          * a passwd(4) entry for login.
2028          */
2029         if (!interactive && !failsafe) {
2030                 if (zone_get_user_cmd(bh, login, user_cmd,
2031                     sizeof (user_cmd)) == NULL) {
2032                         zerror(gettext("could not get user_cmd for zone %s"),
2033                             zonename);
2034                         brand_close(bh);
2035                         return (1);
2036                 }
2037         }
2038         brand_close(bh);
2039 
2040         if ((new_env = prep_env()) == NULL) {
2041                 zperror(gettext("could not assemble new environment"));
2042                 return (1);
2043         }
2044 
2045         if (!interactive)
2046                 return (noninteractive_login(zonename, user_cmd, zoneid,
2047                     new_args, new_env));
2048 
2049         if (zonecfg_in_alt_root()) {
2050                 zerror(gettext("cannot use interactive login with scratch "
2051                     "zone"));
2052                 return (1);
2053         }
2054 
2055         /*
2056          * Things are more complex in interactive mode; we get the
2057          * master side of the pty, then place the user's terminal into
2058          * raw mode.
2059          */
2060         if (get_master_pty() == -1) {
2061                 zerror(gettext("could not setup master pty device"));
2062                 return (1);
2063         }
2064 
2065         /*
2066          * Compute the "short name" of the pts.  /dev/pts/2 --> pts/2
2067          */
2068         if ((slavename = ptsname(masterfd)) == NULL) {
2069                 zperror(gettext("failed to get name for pseudo-tty"));
2070                 return (1);
2071         }
2072         if (strncmp(slavename, "/dev/", strlen("/dev/")) == 0)
2073                 (void) strlcpy(slaveshortname, slavename + strlen("/dev/"),
2074                     sizeof (slaveshortname));
2075         else
2076                 (void) strlcpy(slaveshortname, slavename,
2077                     sizeof (slaveshortname));
2078 
2079         (void) printf(gettext("[Connected to zone '%s' %s]\n"), zonename,
2080             slaveshortname);
2081 
2082         if (set_tty_rawmode(STDIN_FILENO) == -1) {
2083                 reset_tty();
2084                 zperror(gettext("failed to set stdin pty to raw mode"));
2085                 return (1);
2086         }
2087 
2088         if (prefork_dropprivs() != 0) {
2089                 reset_tty();
2090                 zperror(gettext("could not allocate privilege set"));
2091                 return (1);
2092         }
2093 
2094         /*
2095          * We must mask SIGCLD until after we have coped with the fork
2096          * sufficiently to deal with it; otherwise we can race and receive the
2097          * signal before child_pid has been initialized (yes, this really
2098          * happens).
2099          */
2100         (void) sigset(SIGCLD, sigcld);
2101         (void) sigemptyset(&block_cld);
2102         (void) sigaddset(&block_cld, SIGCLD);
2103         (void) sigprocmask(SIG_BLOCK, &block_cld, NULL);
2104 
2105         /*
2106          * We activate the contract template at the last minute to
2107          * avoid intermediate functions that could be using fork(2)
2108          * internally.
2109          */
2110         if ((tmpl_fd = init_template()) == -1) {
2111                 reset_tty();
2112                 zperror(gettext("could not create contract"));
2113                 return (1);
2114         }
2115 
2116         if ((child_pid = fork()) == -1) {
2117                 (void) ct_tmpl_clear(tmpl_fd);
2118                 reset_tty();
2119                 zperror(gettext("could not fork"));
2120                 return (1);
2121         } else if (child_pid == 0) { /* child process */
2122                 int slavefd, newslave;
2123 
2124                 (void) ct_tmpl_clear(tmpl_fd);
2125                 (void) close(tmpl_fd);
2126 
2127                 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
2128 
2129                 if ((slavefd = init_slave_pty(zoneid, devroot)) == -1)
2130                         return (1);
2131 
2132                 /*
2133                  * Close all fds except for the slave pty.
2134                  */
2135                 (void) fdwalk(close_func, &slavefd);
2136 
2137                 /*
2138                  * Temporarily dup slavefd to stderr; that way if we have
2139                  * to print out that zone_enter failed, the output will
2140                  * have somewhere to go.
2141                  */
2142                 if (slavefd != STDERR_FILENO)
2143                         (void) dup2(slavefd, STDERR_FILENO);
2144 
2145                 if (zone_enter(zoneid) == -1) {
2146                         zerror(gettext("could not enter zone %s: %s"),
2147                             zonename, strerror(errno));
2148                         return (1);
2149                 }
2150 
2151                 if (slavefd != STDERR_FILENO)
2152                         (void) close(STDERR_FILENO);
2153 
2154                 /*
2155                  * We take pains to get this process into a new process
2156                  * group, and subsequently a new session.  In this way,
2157                  * we'll have a session which doesn't yet have a controlling
2158                  * terminal.  When we open the slave, it will become the
2159                  * controlling terminal; no PIDs concerning pgrps or sids
2160                  * will leak inappropriately into the zone.
2161                  */
2162                 (void) setpgrp();
2163 
2164                 /*
2165                  * We need the slave pty to be referenced from the zone's
2166                  * /dev in order to ensure that the devt's, etc are all
2167                  * correct.  Otherwise we break ttyname and the like.
2168                  */
2169                 if ((newslave = open(slavename, O_RDWR)) == -1) {
2170                         (void) close(slavefd);
2171                         return (1);
2172                 }
2173                 (void) close(slavefd);
2174                 slavefd = newslave;
2175 
2176                 /*
2177                  * dup the slave to the various FDs, so that when the
2178                  * spawned process does a write/read it maps to the slave
2179                  * pty.
2180                  */
2181                 (void) dup2(slavefd, STDIN_FILENO);
2182                 (void) dup2(slavefd, STDOUT_FILENO);
2183                 (void) dup2(slavefd, STDERR_FILENO);
2184                 if (slavefd != STDIN_FILENO && slavefd != STDOUT_FILENO &&
2185                     slavefd != STDERR_FILENO) {
2186                         (void) close(slavefd);
2187                 }
2188 
2189                 /*
2190                  * In failsafe mode, we don't use login(1), so don't try
2191                  * setting up a utmpx entry.
2192                  */
2193                 if (!failsafe)
2194                         if (setup_utmpx(slaveshortname) == -1)
2195                                 return (1);
2196 
2197                 /*
2198                  * The child needs to run as root to
2199                  * execute the brand's login program.
2200                  */
2201                 if (setuid(0) == -1) {
2202                         zperror(gettext("insufficient privilege"));
2203                         return (1);
2204                 }
2205 
2206                 (void) execve(new_args[0], new_args, new_env);
2207                 zperror(gettext("exec failure"));
2208                 return (1);
2209         }
2210 
2211         (void) ct_tmpl_clear(tmpl_fd);
2212         (void) close(tmpl_fd);
2213 
2214         /*
2215          * The rest is only for the parent process.
2216          */
2217         (void) sigset(SIGWINCH, sigwinch);
2218 
2219         postfork_dropprivs();
2220 
2221         (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
2222         doio(masterfd, -1, masterfd, -1, -1, B_FALSE);
2223 
2224         reset_tty();
2225         (void) fprintf(stderr,
2226             gettext("\n[Connection to zone '%s' %s closed]\n"), zonename,
2227             slaveshortname);
2228 
2229         if (pollerr != 0) {
2230                 (void) fprintf(stderr, gettext("Error: connection closed due "
2231                     "to unexpected pollevents=0x%x.\n"), pollerr);
2232                 return (1);
2233         }
2234 
2235         return (0);
2236 }